International Transportation | Collection 2022 39 European Friedrich-List-Award 2022 SCIENCE & RESEARCH Effects of travel time VMS on-urban traffic Attitudes to travel times displayed on variable message signs and effect on route choice Variable message sign, Travel time, Route choice, Sensitivity function, Stated preference Expected travel time information on variable message signs (VMS) supports even capacity utilization on the road network. To justify the importance of VMS in urban traffic, general user attitude to VMS was assessed and route choice decisions were investigated through stated preferences. Logit model was used to show how route choices are affected by changing expected travel time, reliability of travel time, and perceived travel time of alternative routes. Applying logistic regression, functions were defined that describe the sensitivity of route change depending on expected travel times. Renáta Bordás A ccording to mobility reports (e.- g. published by Inrix, Tom- Tom), congestion is a key issue in many cities causing bottlenecks and generating high external costs. [1, 2] Variable message signs (VMS) and coordinated traffic signals are traffic management tools that support traffic flows to be led on roads with more free capacity. The novelty of this research is to explore the effects of travel time VMS on traffic in the urban environment as this topic is not thoroughly investigated in scientific papers. Expected travel times displayed on variable message sign make drivers change their planned route to an alternative one with less travel time. By influencing route choice decisions, expected travel time signs shift the uneven capacity utilization on the network towards equilibrium. With utilized capacities, congestion relief can be achieved that goes with less delay, driver stress, fewer stops, and conflicts on roads. To assess the importance of travel time VMS, it was essential to unfold the current system’s ability to influence traffic and to learn about its impact on route choice. Therefore, the general attitude and response of traffic participants to expected travel time VMS were assessed. Urban route choice decisions are interpreted based on the effects of expected travel time, reliability of travel time, and perceived travel time. Furthermore, functions were defined that describe the sensitivity of route change regarding the expected travel times displayed on variable message signs. Finally, subnetwork criteria for further application of the results were specified. This paper reports about the assessment of VMS network impacts and its real potential to be used as a traffic management tool. This own research was made to support a R&D project about traffic management. Survey methods and structure Stated preference methods offer the possibility to identify behavioural responses to choice situations that cannot be observed and measured in real time on the roads. In the present research, a wide range of route choice predictor variables were observed with this practice, that would have not been perceptible on the road network with revealed preference methods. [3] The compiled questionnaire served as a two-step survey of users’ reactions to the VMS. Firstly, a general section aimed at exploring driving habits and common user attitudes to travel time VMS. In the second part, respondents were required to state their preferences in discrete choices and hypothetical situations regarding route choices. In the discrete choice experiment (DCE), respondents expressed their preferences based on route attributes as respondents were required to choose one of a set of product cards that referred to traffic conditions of different routes of the same destination. According to the model, the respondents associate different degrees of utility to attributes and levels, the sum of which determines which route is more beneficial to them. [4] The second part of stated preference research carried corresponds to contingent valuation, where respondents were asked how much extra travel time they tolerate to stay on their planned route and how much extra travel time make them change their route due to congestion. Questions according to the mentioned survey methods were asked from the target population, that were people who have ever met travel time VMS. Participants attended from Budapest and selected parts of its agglomeration as the sample area. The sample size can be considered reliable. Attitude to travel time variable message sign In the questionnaire, participants evaluated the usefulness and accuracy of the VMS panels on a five-point Likert scale. Overall, results show positive user attitude towards the VMS system. 81 % of respondents considered the VMS useful and 58 % considered them accurate in displaying the expected travel time. Accuracy got less positive reviews because more respondents remained neutral in this case. In the subset of active driver respondents who have ever seen a VMS panel with travel times displayed, 20 % of drivers take the VMS into account during driving. In the subset of those, who regularly drive on routes where VMS panels are deployed, 24 % of drivers consider the travel time information displayed. International Transportation | Collection 2022 40 SCIENCE & RESEARCH European Friedrich-List-Award 2022 Effects of travel time, travel time reliability, and perceived travel time The characteristics of suburban trips and the surrounding network of VMS panels were observed in Budapest to set the levels- of route attributes. The following route attributes were placed on card pairs with fractional factorial design in DCE questions: •• travel time (10 minutes, 15 minutes, 20-minutes), •• travel time reliability (60 %, 90 %, 99 %), •• perceived travel time (few intersections, many intersections). The levels of route attributes were set to be approximately equidistant on the utility function. [5] The reliability of travel time is expressed as follows: with x % reliability, there is x % chance the driver will complete the route within the given travel time and 100−x % chance that it will take more than that. Perceived travel time is expressed in the number of intersections and turns, which can affect the perception of comfort and thus the perceived travel time through decelerations, stops, and higher risk of accidents. Proper encoding of attribute levels was essential for correct results of the logit model that was used to assess the discrete choices. In the logit model the usual terms “success” (1) and “failure” (0), as the outputs of DCE, represent choosing a given card and not choosing that. Odds of success are defined as the ratio of the probability of success (P x ) over the probability of failure (1-P x ). Equation (1) presents, that according to the model the natural logarithm of the odds, that is the logit of probability, is a linear function of the explanatory variables. The parameters of utility functions are estimated by maximum likelihood method which is an iteration procedure. [6] (1) Logit model was run on two data sets. One dataset contains the responses to situations where respondents had to arrive at exact time (e.g., meeting), the other dataset is based on the responses when drivers could arrive at any time to their destination (e.g., shopping). 1,728 observations were used in both datasets analyses, and 4 iterations were needed for the results. At a 5 % significance level, the model is statistically significant. Coefficients were converted to changes in the probability of a card’s choice resulting from a one-unit change in the attribute. A summary of the results with some indicators of model fitting is shown in Table 1. Values underlined are considered statistically significant at a 5 % significance level. Overall, based on these values, it can be concluded for the three attributes that travel time is not a significant factor in route choice between 10 to 20 minutes if the reliability of this information may vary and the driver needs to arrive at exact time to their destination. Apparently, most people usually have 10 minutes of spare time when they need get on time to their destination. Therefore, in the case of an exact arrival time, the reliability of the information and the number of nodes and turns were the determining factors in the decisions. The number of junctions and turns had a negative effect on the probability of choice, presumably uncertainty was associated with this attribute. With the condition of not determined arrival time, changing travel time had the largest impact on card choices, while a decrease in reliability to 90 % did not show a significant change in choices. It can be concluded that respondents were more willing to take risks (in certainty) to get to their destination faster as they had no chance of being late for their destination. Sensitivity of route changing Regarding two illustrative subnetworks of Budapest, respondents stated how much extra travel time (due to congestion) makes them change their chosen route to an alternative one. The input was the increased travel time of the planned route, and the binary output was the options to stay on ∆ Px Travel time Reliability of travel time Number of intersections Goodness of fit from 10 min. to 15 min. from 15 min. to 20 min. from 99% to 90% from 99% to 60% from few to many LR chi2(4) Prob> chi2 Pseudo-R2 arrive at exact time -0,42% -0,43% -12,05% -53,22% -4,60% 375,75 0,0000 0,1569 arrive at any time -14,58% -16,11% 0,69% -11,52% -1,18% 150,07 0,0000 0,0626 Table 1: Effects of route attributes on probability of route choice Figure 1: Sensitivity functions in different route conditions Source: Author International Transportation | Collection 2022 41 European Friedrich-List-Award 2022 SCIENCE & RESEARCH route (0) or change route (1). Equation (2) shows the logistic regression formula by which route change probability was assigned to the minute values of the expected travel times . The function was fitted by maximum likelihood method. (2) The dataset contains only the responses of those who are willing to change their route based on VMS. The diversity of the functions can be seen in Figure 1. Different route conditions that define the sensitivity of route choice (slope of regression functions) are defined in Table 2. The criteria relate to the length of the routes, the information about the alternative route and the similarity of the alternative routes concerned. Considering the criteria above, sensitivity functions can be used for estimating travel time VMS effects on traffic in similar subnetworks to the sample ones in the questionnaires. Furthermore, these results facilitate the deployment design process of the VMS system. Conclusion The importance of travel time VMS has been justified as a positive user attitude was assessed and 24 % of active drivers concerned consider VMS in their route choice. The reliability of the travel time information is of the atmost relevance for drivers in route choice. Drivers tend to react even at small travel time excess with route changing. In case of congestion, drivers are willing to change their route more likely if the alternative route is more similar and expected travel time information is available for the alternative route as well. These statements concern road sections of 10-20-30 minutes in urban and suburban environments. Survey results support determining potential spots on the network where VMS deployment would have significant benefits for traffic participants and operators. Historical VMS data and traffic data with the sensitivity functions made it possible to calculate realized travel time benefits at a given subnetwork of Budapest for a given period. Time benefits realized also proved the significant role of the current system in managing capacity utilization. Travel time VMS system contributes to leading traffic flow from congested paths, thus reducing travel time, stops, conflicts and stress for drivers. Road network operators also have the potential to incorporate prediction in the VMS system for more efficient traffic management. This research provides a basis for the development of traffic management strategies that include congestion forecasting in expected travel time signs. Thus, operators would be able to intervene before congestion occurs or in the very initial phase. This research has shown the extent to which the current VMS system influences traffic and that the reliability of travel time information plays a crucial role in route choice decisions. It is worth continuing to work on this topic by modeling the effects of travel time VMS and creating a strategy with prediction to be able to intervene in the queueing in a very initial phase. ■ equation 1 equation 2 equation 3 equation 4 equation 5 equation 6 equation 7 β 0 -4,4957 -5,9868 -4,8419 -4,1041 -3,7814 -5,0971 -5,3839 β travel time 0,2148 0,5106 0,3488 0,4254 0,3322 0,5048 0,5276 R 2 0,4070 0,3851 0,3845 0,5029 0,0670 0,4076 0,4254 Route conditions for application of coefficients Route of 8-30 minute-drive (uncongested) x x x x x Route of 2-20 minute-drive (uncongested) x x Expected travel time information is available on alternative route x x Heterogeneous characteristics of alternatives x x x x x Homogeneous characteristics of alternatives x x No traffic information about the alternative route x Urban region x x x x Suburban region x x x Table 2: Subnetwork criteria for further application of the equations Renáta Bordás, MSc. Transportation Engineer, Department for Transportation Planning, Főmterv Civil Engineering Ltd., Budapest (HU) bordas.renata@fomterv.hu The research reported in this paper was supported by the market-driven research, development and innovation projects (2019-1.1.1-PIACI-KFI-2019-00330). REFERENCES [1] B. Pishue, P. (2021): INRIX Global Traffic Scorecard. https: / / inrix.com/ (Access: 2022.04.25.) [2] TomTom International BV, Traffic Index 2021. www.tomtom.com/ en_gb/ traffic-index/ ranking/ (Access: 2022.04.25.) [3] Kroes, E. P.; Sheldon, R. J. (1988): Stated preference methods In: Journal of Transport Economics and Policy (Vol. 22), No. 1, pp. 12-25 [4] Baji, P. (2012): A diszkrét választás módszere. In: Statisztikai Szemle (Vol. 90), No. 10, pp. 943-963 [5] Daly, A.; Dekker, T.; Hess, S. (2016): Dummy coding vs effects coding for categorical variables: Clarifications and extensions. In: Journal of Choice Modelling (Vol. 21), pp. 36-41 [6] Hajdú, O. (2016): Ökonometria, Budapest, BME Üzleti Tudományok Intézet pp. 89-97